Since the company’s founding in 2018, Voiceflow engineers have worked steadily to create Machine Learning models for building conversational AI agents. “We help our users create conversations, using AI under the hood in a no-code way so everyone can jump into the tool and create conversations with customers,” says Xavier (Xavi) Portilla Edo, Infrastructure Team Lead at Voiceflow.
Voiceflow works with customers to train their chatbots on custom data sets using Voiceflow’s own knowledge base. The value of the company’s datasets means that engineers like Xavi are cautious about the third-party solutions they bring into the business. “If our customers prefer, we can deploy a whole private cloud for them,” Xavi explains. It’s an innovative approach, but Voiceflow must ensure that third-party solutions are not using the company’s data, or the data of Voiceflow customers, to train the vendor’s own AI solutions.
Therefore, data privacy is critical, as is the DevOps model, which Voiceflow has embraced even beyond engineering.
“At Voiceflow, DevOps is not something that’s managed just by the DevOps team,” Xavi says. “It’s more a philosophy. It’s established in every team – from developers to data scientists.”
Doing away with heavy CI/CD maintenance and management
In the past, implementing CI/CD at Voiceflow was challenging. Since just about everyone was encouraged to adopt DevOps, there were long training and ramp-up times for non-engineers.
“We had to host everything on our end,” Xavi says. “And we had to use tools that were not mature enough. Once we began growing and scaling, we had to switch to a cloud CI/CD solution.”
Keeping in mind the company needed a cloud solution that was easy to learn to use while also easily integrating AI and ML, Voiceflow chose CircleCI.
“CircleCI was super-easy to set up – in about one day, we had one whole workflow running,” Xavi says. “The maturity and the robustness of the tool was perfect and fits well with our needs.”
The other CI/CD solutions that Voiceflow considered, such as AWS and GitHub Actions, were either very complicated to use or didn’t include as many developer-friendly features. “CircleCI has been around for a while, and there are many things like the continuous pipelines that some of the other solutions don’t have,” Xavi says.
Organic adoption of CircleCI thanks to ease of use
True to the expectations of Voiceflow developers, CircleCI adoption has spread far in the company.
“The entire company is using CircleCI,” Xavi says. “The DevOps team rolled out CircleCI for the front end, back end, platform engineering, and the developer experience. The data scientists use it to create our models and update the data warehouse. With the machine learning team, it’s the same – they use CircleCI to build and deploy their models everywhere.”
CircleCI allows Voiceflow teams to create several different types of machine learning models. “We build our own models, and that’s our regular build in CircleCI,” Xavi explains. “When we’re building super-small applications for machine-learning inference that are running on our clusters, it’s easy to achieve because of all the processes we’re using in CircleCI.”
More frequent testing now possible
The rise of generative AI is putting pressure on developers to test more frequently – a strategy that CircleCI supports. “Every single application should be up and running 100 percent, which is why we need effective CI/CD,” Xavi says.
Today, there are far fewer barriers to the development of chatbots and AI/ML solutions at Voiceflow – which, as Xavi explains, helps the company create better products for its end users.
“When you want to create a bot or an assistant, it should be just a matter of writing a description,” says Xavi. “Everything we build and deploy is created with CircleCI – and it’s helping us build more models and machine learning applications.”
“Everything we build and deploy is created with CircleCI – and it’s helping us build more models and machine learning applications.”
Xavier Portilla Edo
Infrastructure Team Lead at Voiceflow
Benefits of using CircleCI:
- Adoption by multiple teams for additional use cases
- Reduced management and maintenance
- Protection for valuable data
Voiceflow is building the world’s best platform and community for the creation of AI agents, for any use case. Voiceflow plugs on-top of any existing NLU platform, LLM, or technology, allowing teams to supercharge their design and collaboration capabilities without any costly vendor replacements or technology changes.